Anbu HuangGenerative AI & Creative Systems

Research · Build · Share

Anbu Huang

Researching generative intelligence. Building creative systems.

Research Scientist, Author, AI Enthusiast · THU

Shenzhen skyline beside the bay
Anbu Huang
Shenzhen, China

Research focus

Read my writing

I study Generative Models (Diffusion/Flow-Matching) and VLM Foundation Models, for which you can find more in my blog post.

01 / Generative modeling

Efficient and controllable generation

I study diffusion and flow-matching models, focusing on generation quality, sampling efficiency, and controllability. My work explores distillation, guidance, and solver-aware sampling to improve training and inference, with the goal of producing high-quality results with less computation and more precise control.

02 / Vision–language modeling

Unified understanding and generation

I explore native vision–language architectures that bring understanding and generation into a shared model. My goal is to connect visual perception, reasoning, and content creation, so that models can interpret what they see and use that understanding to guide what they generate.

My earlier research focused on recommender systems, federated learning, and AI safety, with publications at conferences including ICLR and AAAI.

Selected project / Code2X

Genesis

One idea. Three ways to create.

A video creation studio connecting code-driven animation, model-generated storytelling, and hybrid exploration. From precise explanations to expressive visual stories.

Open the studio
Planetary orbits / Manim animationFull demo · 60 s · Audio
Available · Programmatic video

Make the abstract
visible.

Turn a concept into an editable, narrated Manim animation.

  1. Plan the explanationPlan a storyboard from your topic, duration and style, with controlled geometry, equations and motion.
  2. Write, voice & renderGenerate Manim code and Chinese or English AI narration; render each shot and repair errors.
  3. Align & exportAlign animation, narration and subtitles. Export MP4, WAV, SRT / VTT, source code and timeline.
Explore code-driven creation
Art class / Generative videoFull demo · 15 s · Audio
Available · Generative video

From a brief
to a visual story.

Shape one story across multiple generated shots.

  1. Define the story worldPlan 2–8 shots from your brief, with visual references for characters, scenes and overall style.
  2. Generate & reviewOptionally review sampled frames, refine prompts and select takes. Guide continuity with visual memory and scene state.
  3. Assemble the storyNormalize and join clips on one timeline. Inspect the storyboard and production records.

Refinement and continuity are independent; results vary by model.

Explore model-generated stories
STRUCTUREATMOSPHERE
Interactive concept illustrationHybrid preview
Preview · Research direction

Precise by design.
Expressive by nature.

Use code to define structure and motion, then explore model-driven atmosphere and texture. A creative direction for bringing both approaches into one workflow.

  • Structure to imagery
  • Animation to texture
  • Assets to scenes
Program structureGenerative appearance

The hybrid workspace is a UI preview; generation is not enabled yet.

Explore the hybrid preview

The web workspace connects to your own local engine and configured models for generation. Genesis is the studio interface of my Code2X project.

Web project on GitHub